Information flow recommendation method, apparatus, and computer program product

Abstract

The present disclosure provides an information flow recommendation method and device, electronic equipment, storage medium and program product, relates to the technical field of artificial intelligence, in particular to the technical field of evolutionary strategy. The specific implementation scheme is: obtaining feature information of a first user in an information flow recommendation scene; determining, by a multi-factor fusion parameter network, a first weight corresponding to each factor in a factor set according to the feature information, wherein the factors in the factor set represent index information to be considered in the information flow recommendation process; determining, by a gating screening network, a second weight corresponding to each factor in the factor set according to the feature information; determining a target factor in the factor set suitable for the first user in the information flow recommendation scene according to the first weight and the second weight; and determining and pushing a recommendation result corresponding to the first user in the information flow recommendation scene to the first user according to the target factor. The present disclosure improves the accuracy of the recommendation result.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, specifically to the field of evolutionary strategy technology, and in particular to information flow recommendation methods, devices, model training methods, devices, electronic devices, storage media, and computer program products, which can be used in information flow recommendation scenarios. Background Technology

[0002] Unlike advertising, news feed recommendations not only focus on the click-through rate of resources but also integrate a series of experience metrics such as reading time, the diversity of displayed resources, user likes, and shares as a comprehensive recommendation indicator. Although the number of target factors being integrated is increasing, different factors have their applicable scenario limitations. For example, for new user models, the primary task is to activate and acquire new users; goals such as reading time and diversity are not the system's main focus. How to adaptively select factors based on the user's context is a common problem in news feed recommendation systems. Summary of the Invention

[0003] This disclosure provides an information flow recommendation method, apparatus, model training method, apparatus, electronic device, storage medium, and computer program product.

[0004] According to the first aspect, an information flow recommendation method is provided, comprising: acquiring feature information of a first user in an information flow recommendation scenario; determining a first weight corresponding to each factor in a factor set based on the feature information through a multi-factor fusion parameter network, wherein the factors in the factor set represent the indicator information to be considered in the information flow recommendation process; determining a second weight corresponding to each factor in the factor set based on the feature information through a gating screening network; determining a target factor in the factor set suitable for the first user in the information flow recommendation scenario based on the first weight and the second weight; and determining and pushing the recommendation result corresponding to the first user in the information flow recommendation scenario to the first user based on the target factor.

[0005] According to the second aspect, a model training method is provided, comprising: acquiring feature information of a second user in an information flow recommendation scenario; determining the first weight corresponding to each factor in the factor set based on the feature information through an initial multi-factor fusion parameter network, wherein the factors in the factor set represent the indicator information to be considered in the information flow recommendation process; determining the second weight corresponding to each factor in the factor set based on the feature information through an initial gating filter network; determining the target factor in the factor set suitable for the second user in the information flow recommendation scenario based on the first weight and the second weight; determining the recommendation result corresponding to the second user in the information flow recommendation scenario based on the target factor; and adopting an evolutionary strategy to adjust the parameters of the initial multi-factor fusion parameter network and the initial gating filter network based on the feedback information of the second user on the recommendation result, so as to obtain the trained multi-factor fusion parameter network and the gating filter network.

[0006] According to a third aspect, an information flow recommendation device is provided, comprising: a first acquisition unit configured to acquire feature information of a first user in an information flow recommendation scenario; a first determination unit configured to determine a first weight corresponding to each factor in a factor set based on the feature information through a multi-factor fusion parameter network, wherein the factors in the factor set represent indicator information to be considered in the information flow recommendation process; a second determination unit configured to determine a second weight corresponding to each factor in the factor set based on the feature information through a gating filtering network; a third determination unit configured to determine a target factor in the factor set suitable for the first user in the information flow recommendation scenario based on the first weight and the second weight; and a recommendation unit configured to determine and push recommendation results corresponding to the first user in the information flow recommendation scenario based on the target factor.

[0007] According to the fourth aspect, a model training device is provided, comprising: a second acquisition unit configured to acquire feature information of a second user in an information flow recommendation scenario; a fourth determination unit configured to determine a first weight corresponding to each factor in a factor set based on the feature information through an initial multi-factor fusion parameter network, wherein the factors in the factor set represent the indicator information to be considered in the information flow recommendation process; a fifth determination unit configured to determine a second weight corresponding to each factor in the factor set based on the feature information through an initial gating filter network; a sixth determination unit configured to determine a target factor in the factor set suitable for the second user in the information flow recommendation scenario based on the first weight and the second weight; a seventh determination unit configured to determine the recommendation result corresponding to the second user in the information flow recommendation scenario based on the target factor; and a training unit configured to adopt an evolutionary strategy to adjust the parameters of the initial multi-factor fusion parameter network and the initial gating filter network based on the feedback information of the second user on the recommendation result, so as to obtain the trained multi-factor fusion parameter network and the gating filter network.

[0008] According to a fifth aspect, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a method as described in any implementation of the first or second aspect.

[0009] According to a sixth aspect, a non-transitory computer-readable storage medium is provided that stores computer instructions for causing a computer to perform the method described in any implementation of the first or second aspect.

[0010] According to the seventh aspect, a computer program product is provided, comprising: a computer program that, when executed by a processor, implements the method as described in either the first or second aspect.

[0011] According to the technology disclosed herein, an information flow recommendation method is provided. In the information flow recommendation scenario, a first weight of each factor corresponding to the user is determined by a multi-factor fusion parameter network, and a second weight of each factor is determined by a gating screening network. Based on the first weight and the second weight, the target factors suitable for the user in the information flow recommendation scenario are accurately determined, and information flow recommendation is performed, thereby improving the accuracy of the recommendation results.

[0012] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0013] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0014] Figure 1 This is an exemplary system architecture diagram that can be applied to an embodiment of this disclosure;

[0015] Figure 2 This is a flowchart of an embodiment of the information flow recommendation method according to this disclosure;

[0016] Figure 3 This is a schematic diagram illustrating an application scenario of the information flow recommendation method according to this embodiment;

[0017] Figure 4 This is a flowchart of yet another embodiment of the information flow recommendation method according to the present disclosure;

[0018] Figure 5 This is a flowchart of an embodiment of the model training method according to the present disclosure;

[0019] Figure 6 This is a structural diagram of an embodiment of the information flow recommendation device according to the present disclosure;

[0020] Figure 7 This is a structural diagram of one embodiment of the model training apparatus according to the present disclosure;

[0021] Figure 8 This is a schematic diagram of the structure of a computer system suitable for implementing the embodiments of the present disclosure. Detailed Implementation

[0022] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0023] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0024] Figure 1 An exemplary architecture 100 is shown that the information flow recommendation method and apparatus, and the model training method and apparatus of this disclosure can be applied.

[0025] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. The communication connections between terminal devices 101, 102, and 103 form a network topology. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0026] Terminal devices 101, 102, and 103 can be hardware or software that supports network connectivity for data interaction and processing. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices that support network connectivity, information acquisition, interaction, display, and processing functions, including but not limited to smartphones, tablets, e-book readers, laptops, and desktop computers. When terminal devices 101, 102, and 103 are software, they can be installed in the aforementioned electronic devices. They can be implemented as, for example, multiple software programs or software modules to provide distributed services, or as a single software program or software module. No specific limitations are imposed here.

[0027] Server 105 can be a server providing various services. For example, based on the feature information of users corresponding to terminal devices 101, 102, and 103 in an information flow recommendation scenario, it determines the first weight of each factor corresponding to the user through a multi-factor fusion parameter network and the second weight of each factor through a gating filter network, so as to accurately determine the target factors suitable for the user based on the first and second weights, and perform information flow recommendation as a backend processing server. Another example is a backend processing server that trains a multi-factor fusion parameter network and a gating filter network based on an evolutionary strategy, using feedback information on recommendation results provided by terminal devices 101, 102, and 103. As an example, server 105 can be a cloud server.

[0028] It should be noted that a server can be either hardware or software. When the server is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When the server is software, it can be implemented as multiple software programs or software modules (such as software programs or software modules used to provide distributed services), or as a single software program or software module. No specific limitations are made here.

[0029] It should also be noted that the information flow recommendation method and model training method provided in the embodiments of this disclosure can be executed by a server, by a terminal device, or by a combination of both. Accordingly, the various parts (e.g., units) included in the information flow recommendation device and model training device can be all located in the server, all located in the terminal device, or located in both the server and the terminal device respectively.

[0030] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included. When the electronic devices on which the information flow recommendation method and model training method run do not require data transmission with other electronic devices, the system architecture may only include the electronic devices (e.g., servers or terminal devices) on which the information flow recommendation method and model training method run.

[0031] Please refer to Figure 2 , Figure 2 A flowchart of an information flow recommendation method provided in this disclosure embodiment, wherein process 200 includes the following steps:

[0032] Step 201: Obtain the feature information of the first user in the information flow recommendation scenario.

[0033] In this embodiment, the execution body of the information flow recommendation method (e.g., Figure 1The terminal device or server in the process can obtain the first user's feature information in the information flow recommendation scenario from a remote location or from a local location based on a wired network connection or a wireless network connection.

[0034] The first user is the user to be recommended information in the feed. The information feed recommendation scenario can be any type of information feed. For example, in news applications, the information feed recommendation scenario is to determine the news information feed that the user is interested in; in video applications, the information feed recommendation scenario is to determine the video information feed that the user is interested in.

[0035] The first user's characteristic information in the information feed recommendation scenario includes the user's characteristic information and the scenario characteristic information of the information feed recommendation scenario. As an example, the user characteristic information includes user activity level, age, gender, average daily product usage time, and number of uses; the scenario characteristic information includes refresh status, number of refreshes, and refresh time.

[0036] Step 202: Determine the first weight of each factor in the factor set based on the feature information through a multi-factor fusion parameter network.

[0037] In this embodiment, the aforementioned execution entity can determine the first weight corresponding to each factor in the factor set based on feature information through a multi-factor fusion parameter network.

[0038] The factors in the factor set represent the metrics that need to be considered in the information flow recommendation process. For example, the factor set includes factors such as reading time, the diversity of displayed resources, user likes, and shares.

[0039] In this implementation, to improve the targeting of information flow recommendations, different factor sets can be determined for different information flow recommendation scenarios. The factor set corresponding to each information flow recommendation scenario includes multiple factors specific to that scenario.

[0040] As an example, a multi-factor fusion parameter network includes multiple tower networks, each corresponding to a factor in the factor set. At the bottom layer of the multi-factor fusion parameter network, the multiple tower networks share feature information and are used to output the first weight of their respective factors. Each tower network can be a neural network, including but not limited to convolutional neural networks, recurrent neural networks, and other network models.

[0041] Step 203: Using a gating screening network, determine the second weight corresponding to each factor in the factor set based on the feature information.

[0042] In this embodiment, the aforementioned execution entity can determine the second weight corresponding to each factor in the factor set based on feature information through a gating screening network.

[0043] Gated screening networks can be implemented based on gated recurrent neural networks. As an example, the aforementioned execution entity can determine the network structure of the gated screening network based on the number of factors included in the factor set, ensuring that the number of outputs of the gated screening network matches the number of factors included in the factor set, and that there is a one-to-one correspondence between the factors in the factor set and the outputs of the gated screening network. The multiple outputs of the gated screening network represent the second weights corresponding to each factor in the factor set.

[0044] Step 204: Determine the target factor in the factor set that is suitable for the first user in the information flow recommendation scenario, based on the first weight and the second weight.

[0045] In this embodiment, the aforementioned execution entity can determine the target factor of the factor set that is suitable for the first user in the information flow recommendation scenario based on the first weight and the second weight.

[0046] As an example, for each factor in the factor set, the aforementioned execution entity can determine the first weight and the second weight corresponding to the factor, and determine the total weight by summation, weighted summation, or other methods; then sort the total weights, and determine the first preset number of factors as target factors, or determine the factors with a total weight greater than a preset value as target factors.

[0047] Step 205: Based on the target factor, determine and push the recommendation results corresponding to the first user in the information flow recommendation scenario to the first user.

[0048] In this embodiment, the aforementioned execution entity can determine and push the recommendation results corresponding to the first user in the information flow recommendation scenario to the first user based on the target factor.

[0049] As an example, for each determined target factor, the aforementioned execution entity can first obtain the total weight of the target factor based on the first weight and the second weight corresponding to the target factor; then, based on the corresponding target factor and the total weight, obtain the weighting term corresponding to each target factor, and combine the weighting terms of each target factor to obtain the multi-target factor fusion formula.

[0050] After obtaining the multi-objective factor fusion formula, the recommendation ranking score of the content to be ranked in the preset set of content to be recommended can be determined by the multi-objective factor fusion formula; based on the recommendation ranking score, the content to be ranked in the set of content to be ranked is sorted, and the preset number of content to be ranked at the top is used as the recommendation result corresponding to the first user and pushed to the first user.

[0051] See also Figure 3 , Figure 3 This is a schematic diagram 300 illustrating an application scenario of the information flow recommendation method according to this embodiment. Figure 3In the application scenario, user 301 sends a launch command to a short video application via terminal device 302. Server 303 first obtains the feature information of user 301 in the short video information flow recommendation scenario based on the launch command; then, through a multi-factor fusion parameter network 304, it determines the first weight 305 corresponding to each factor in the factor set based on the feature information, wherein the factors in the factor set represent the indicator information to be considered in the information flow recommendation process; then, through a gating filtering network 306, it determines the second weight 307 corresponding to each factor in the factor set based on the feature information; then, based on the first weight 305 and the second weight 307, it determines the target factor 308 of the factor set that is suitable for the first user in the information flow recommendation scenario; based on the target factor 308, it determines and pushes the recommendation result 309 corresponding to the first user in the information flow recommendation scenario to user 301.

[0052] This embodiment provides an information flow recommendation method. In the information flow recommendation scenario, a multi-factor fusion parameter network is used to determine the first weight of each factor corresponding to the user, and a gating screening network is used to determine the second weight of each factor. Based on the first and second weights, the target factors suitable for the user are accurately determined for information flow recommendation, thereby improving the accuracy of the recommendation results.

[0053] In some optional implementations of this embodiment, the execution entity can perform step 203 as follows:

[0054] First, the initial second weights of each factor in the factor set are determined by a gating network based on feature information. Then, the initial second weights of each factor are converted to 0 / 1 by a preset activation function to obtain the second weights of each factor.

[0055] As an example, the preset activation function of the top layer of the gated filtering network has a range of 0 / 1, thus cleverly transforming the continuous value problem into a 0 / 1 problem. For example, its preset activation function could be:

[0056]

[0057] In this implementation, for each factor in the factor set, the second weight output by the gating screening network is 0 / 1, thus cleverly transforming the continuous value problem into a 0 / 1 problem, improving the efficiency and convenience of the target factor determination process.

[0058] In some optional implementations of this embodiment, the execution entity can perform step 204 as follows:

[0059] First, multiply the first weight and second weight of each factor in the factor set to obtain a weight product; then, based on the weight product of each factor in the factor set, determine the target factor in the factor set that is suitable for the first user in the information flow recommendation scenario.

[0060] When the second weight output by the gating network is 0 / 1, it is easy to determine that the weight product of each factor is the first weight of that factor or zero; then, factors with zero weight product are removed, and factors with non-zero weight product are retained to obtain the target factor.

[0061] In this implementation, the target factor is determined based on the weight product of the first and second weights corresponding to each factor, which further improves the convenience and accuracy of the target factor determination process.

[0062] In some optional implementations of this embodiment, the above-mentioned execution entity may also perform the following operations: First, obtain feedback information from the first user regarding the recommendation results.

[0063] The feedback information can be the user's reaction to the information stream in the recommendation results after receiving them. For example, feedback information includes whether the user clicked, viewed, or performed interactive actions such as liking or commenting.

[0064] Second, an evolutionary strategy is adopted to adjust the parameters of the multi-factor fusion parameter network and the gated filtering network based on feedback information, so as to execute the subsequent user recommendation task in the information flow recommendation scenario through the adjusted multi-factor fusion parameter network and the gated filtering network.

[0065] Evolutionary strategy algorithms refer to algorithms based on evolutionary theory. These algorithms explore and seek parameter perturbations that maximize the overall reward of the multi-factor fusion parameter network and the gated screening network. Specifically, based on feedback information and a preset reward function, the reward values ​​of the parameters of the multi-factor fusion parameter network and the gated screening network are determined. The principle of maximizing reward values ​​guides the adjustment process of the parameters of the multi-factor fusion parameter network and the gated screening network. The algorithm iterates according to a preset number of iterations based on the preset evolutionary strategy algorithm, generating a new round of parameters for the multi-factor fusion parameter network and the gated screening network that satisfy a Gaussian distribution of the mean and variance of each parameter.

[0066] The subsequent user recommendation task will be performed using a multi-factor fusion parameter network with adjusted parameters and a gated screening network.

[0067] In this implementation, the aforementioned execution entity uses an evolutionary strategy to adjust the multi-factor fusion parameter network and the gated screening network during their application, which can continuously improve the accuracy of the multi-factor fusion parameter network and the gated screening network.

[0068] Continue to refer to Figure 4 The illustration shows a schematic flow 400 of another embodiment of the information flow recommendation method according to the present disclosure, including the following steps:

[0069] Step 401: Obtain the feature information of the first user in the information flow recommendation scenario.

[0070] Step 402: Determine the first weight of each factor in the factor set based on the feature information through a multi-factor fusion parameter network.

[0071] Among them, the factors in the factor set represent the indicator information that needs to be considered in the information flow recommendation process.

[0072] Step 403: Using a gated screening network, determine the initial second weights corresponding to each factor in the factor set based on the feature information.

[0073] Step 404: Convert the initial second weights of each factor to 0 / 1 using a preset activation function to obtain the second weights of each factor.

[0074] Step 405: Multiply the first weight and the second weight corresponding to each factor in the factor set to obtain the weight product.

[0075] Step 406: Determine the target factor in the factor set that is suitable for the first user in the information flow recommendation scenario based on the weight product of each factor in the factor set.

[0076] Step 407: Based on the target factor, determine and push the recommendation results corresponding to the first user in the information flow recommendation scenario to the first user.

[0077] As can be seen from this embodiment, with Figure 2 Compared with the corresponding embodiments, the flow 400 of the information flow recommendation method in this embodiment specifically illustrates the process of determining the second weight and the process of determining the target factor, which further improves the efficiency and convenience of the target factor determination process and improves the accuracy of the recommendation results.

[0078] Continue to refer to Figure 5 The illustration shows a schematic flow 500 of an embodiment of the model training method according to the present disclosure, including the following steps:

[0079] Step 501: Obtain the feature information of the second user in the information flow recommendation scenario.

[0080] In this embodiment, the execution body of the model training method (e.g., Figure 1 The terminal device or server in the process can obtain the feature information of a second user in the information flow recommendation scenario from a remote location or from a local location based on a wired network connection or a wireless network connection.

[0081] In this context, the second user refers to the user to be recommended information during the training of the initial multi-factor fusion parameter network and the initial gating screening network. Multiple second users are typically involved in the model training process. For each second user, the training process shown in steps 501-506 can be executed.

[0082] Information feed recommendation scenarios can be various types of information feeds. For example, in news applications, the information feed recommendation scenario is a news information feed that identifies the user; in video applications, the information feed recommendation scenario is a video information feed that identifies the user.

[0083] The second user's characteristic information in the information feed recommendation scenario includes the user's characteristic information and the scenario characteristic information of the information feed recommendation scenario. As an example, the user characteristic information includes user activity level, age, gender, average daily product usage time, and number of uses; the scenario characteristic information includes refresh status, number of refreshes, and refresh time.

[0084] Step 502: Determine the first weight of each factor in the factor set based on the feature information through the initial multi-factor fusion parameter network.

[0085] In this embodiment, the aforementioned execution entity can determine the first weight corresponding to each factor in the factor set based on feature information through an initial multi-factor fusion parameter network.

[0086] The factors in the factor set represent the metrics that need to be considered in the information flow recommendation process. For example, the factor set includes factors such as reading time, the diversity of displayed resources, user likes, and shares.

[0087] In this implementation, to improve the targeting of information flow recommendations, different factor sets can be determined for different information flow recommendation scenarios. The factor set corresponding to each information flow recommendation scenario includes multiple factors specific to that scenario.

[0088] As an example, the initial multi-factor fusion parameter network includes multiple tower networks, each corresponding to a factor in the factor set. At the bottom layer of the initial multi-factor fusion parameter network, the multiple tower networks share feature information and are used to output the first weight of their respective factors. Each tower network can be a neural network, including but not limited to convolutional neural networks, recurrent neural networks, and other network models.

[0089] Step 503: Using the initial gating screening network, determine the second weight corresponding to each factor in the factor set based on the feature information.

[0090] In this embodiment, the aforementioned execution entity can determine the second weight corresponding to each factor in the factor set based on feature information through an initial gating screening network.

[0091] The initial gated screening network can be implemented based on a gated recurrent neural network. As an example, the aforementioned execution entity can determine the network structure of the initial gated screening network based on the number of factors included in the factor set, ensuring that the number of outputs of the initial gated screening network matches the number of factors included in the factor set, and that there is a one-to-one correspondence between the factors in the factor set and the outputs of the initial gated screening network. The multiple outputs of the initial gated screening network represent the second weights corresponding to each factor in the factor set.

[0092] Step 504: Determine the target factor in the factor set that is suitable for the information flow recommendation scenario for the second user, based on the first weight and the second weight.

[0093] In this embodiment, the aforementioned execution entity can determine the target factor of the second user in the factor set that is suitable for the information flow recommendation scenario based on the first weight and the second weight.

[0094] As an example, for each factor in the factor set, the aforementioned execution entity can determine the first weight and the second weight corresponding to the factor, and determine the total weight by summation, weighted summation, or other methods; then sort the total weights, and determine the first preset number of factors as target factors, or determine the factors with a total weight greater than a preset value as target factors.

[0095] Step 505: Determine the recommendation result for the second user in the information flow recommendation scenario based on the target factor.

[0096] In this embodiment, the aforementioned execution entity can determine the recommendation result corresponding to the second user in the information flow recommendation scenario based on the target factor.

[0097] As an example, for each determined target factor, the aforementioned execution entity can first obtain the total weight of the target factor based on the first weight and the second weight corresponding to the target factor; then, based on the corresponding target factor and the total weight, obtain the weighting term corresponding to each target factor, and combine the weighting terms of each target factor to obtain the multi-target factor fusion formula.

[0098] After obtaining the multi-objective factor fusion formula, the recommendation ranking score of the content to be ranked in the preset set of content to be recommended can be determined by the multi-objective factor fusion formula; based on the recommendation ranking score, the content to be ranked in the set of content to be ranked is sorted, and the preset number of content to be ranked at the top is used as the recommendation result for the second user and pushed to the second user.

[0099] Step 506: Using an evolutionary strategy, adjust the parameters of the initial multi-factor fusion parameter network and the initial gating filter network based on the feedback information from the second user regarding the recommendation results, so as to obtain the trained multi-factor fusion parameter network and gating filter network.

[0100] In this embodiment, the aforementioned execution entity can adopt an evolutionary strategy to adjust the parameters of the initial multi-factor fusion parameter network and the initial gating filter network based on the feedback information from the second user regarding the recommendation results, so as to obtain the trained multi-factor fusion parameter network and gating filter network.

[0101] Specifically, based on feedback information and a preset reward function, the parameters of the initial multi-factor fusion parameter network and the parameters of the initial gated screening network are determined. Based on the principle of maximizing the reward value, the adjustment process of the parameters of the initial multi-factor fusion parameter network and the initial gated screening network is guided. Based on a preset evolutionary strategy algorithm, the process iterates according to a preset number of iterations to generate new rounds of multi-factor fusion parameter network and gated screening network parameters that satisfy a Gaussian distribution of the mean and variance of each parameter. The adjusted multi-factor fusion parameter network and gated screening network are then used as the initial multi-factor fusion parameter network and the initial gated screening network for the next round of training.

[0102] By iteratively executing the above training operations, in response to reaching the preset termination condition, the trained multi-factor fusion parameter network and gating screening network are obtained. The preset termination condition could be, for example, the number of iterations exceeding a preset threshold, or the training time exceeding a preset time threshold.

[0103] In this embodiment, the selection of target factors based on the gating screening network effectively improves the evolutionary efficiency of the evolutionary strategy, and can also trigger the automatic selection of target factors that are beneficial to the whole from the perspective of global optimization.

[0104] In some optional implementations of this embodiment, the execution entity can perform step 503 as follows:

[0105] First, the initial second weights of each factor in the factor set are determined based on the feature information through an initial gating screening network. Then, the initial second weights of each factor are converted to 0 / 1 through a preset activation function to obtain the second weights of each factor.

[0106] As an example, the initial gated filtering network's topmost activation function has a range of 0 / 1, cleverly transforming the continuous value problem into a 0 / 1 problem. For example, its default activation function could be:

[0107]

[0108] In this implementation, for each factor in the factor set, the second weight output by the initial gating screening network is 0 / 1, thus cleverly transforming the continuous value problem into a 0 / 1 problem, further improving the evolutionary efficiency of the evolutionary strategy.

[0109] In some optional implementations of this embodiment, the execution entity can perform step 504 in the following manner:

[0110] First, multiply the first weight and the second weight corresponding to each factor in the factor set to obtain the weight product; then, based on the weight product corresponding to each factor in the factor set, determine the target factor in the factor set that is suitable for the second user in the information flow recommendation scenario.

[0111] When the second weight output by the initial gating network is 0 / 1, it is easy to determine that the weight product of each factor is the first weight of that factor or zero; then, factors with a weight product of zero are removed, and factors with a weight product of non-zero are retained to obtain the target factor.

[0112] In this implementation example, the target factor is determined based on the weight product of the first and second weights corresponding to each factor, which further improves the convenience and accuracy of the target factor determination process during model training.

[0113] Continue to refer to Figure 6 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of an information flow recommendation device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0114] like Figure 6As shown, the information flow recommendation device 600 includes: a first acquisition unit 601, configured to acquire feature information of a first user in an information flow recommendation scenario; a first determination unit 602, configured to determine the first weight corresponding to each factor in a factor set based on the feature information through a multi-factor fusion parameter network, wherein the factors in the factor set represent the indicator information to be considered in the information flow recommendation process; a second determination unit 603, configured to determine the second weight corresponding to each factor in the factor set based on the feature information through a gating screening network; a third determination unit 604, configured to determine the target factor in the factor set applicable to the first user in the information flow recommendation scenario based on the first weight and the second weight; and a recommendation unit 605, configured to determine and push the recommendation result corresponding to the first user in the information flow recommendation scenario to the first user based on the target factor.

[0115] In some optional implementations of this embodiment, the second determining unit 603 is further configured to: determine the initial second weights corresponding to each factor in the factor set based on feature information through a gating screening network; and convert the initial second weights corresponding to each factor into 0 / 1 through a preset activation function to obtain the second weights corresponding to each factor.

[0116] In some optional implementations of this embodiment, the third determining unit 604 is further configured to: multiply the first weight and the second weight corresponding to each factor in the factor set to obtain a weight product; and determine the target factor in the factor set that is suitable for the first user in the information flow recommendation scenario based on the weight product corresponding to each factor in the factor set.

[0117] In some optional implementations of this embodiment, the above-mentioned apparatus further includes: a feedback unit (not shown in the figure), configured to obtain feedback information from the first user regarding the recommendation results; and an evolution unit (not shown in the figure), configured to adopt an evolution strategy to adjust the parameters of the multi-factor fusion parameter network and the gated filtering network according to the feedback information, so as to execute the recommendation task for subsequent users in the information flow recommendation scenario through the adjusted multi-factor fusion parameter network and the gated filtering network.

[0118] In this embodiment, an information flow recommendation device is provided. In the information flow recommendation scenario, a first weight of each factor corresponding to the user is determined by a multi-factor fusion parameter network, and a second weight of each factor is determined by a gating screening network. Based on the first weight and the second weight, the target factors suitable for the user are accurately determined, and information flow recommendations are performed, thereby improving the accuracy of the recommendation results.

[0119] Continue to refer to Figure 7 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of a model training device, which is similar to... Figure 5Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0120] like Figure 7 As shown, the model training device 700 includes: a second acquisition unit 701, configured to acquire feature information of a second user in an information flow recommendation scenario; a fourth determination unit 702, configured to determine the first weight corresponding to each factor in the factor set based on the feature information through an initial multi-factor fusion parameter network, wherein the factors in the factor set represent the indicator information to be considered in the information flow recommendation process; a fifth determination unit 703, configured to determine the second weight corresponding to each factor in the factor set based on the feature information through an initial gating filter network; a sixth determination unit 704, configured to determine the target factor in the factor set suitable for the second user in the information flow recommendation scenario based on the first weight and the second weight; a seventh determination unit 705, configured to determine the recommendation result corresponding to the second user in the information flow recommendation scenario based on the target factor; and a training unit 706, configured to adopt an evolutionary strategy to adjust the parameters of the initial multi-factor fusion parameter network and the initial gating filter network based on the feedback information of the second user on the recommendation result, so as to obtain the trained multi-factor fusion parameter network and the gating filter network.

[0121] In some optional implementations of this embodiment, the fifth determining unit 703 is further configured to: determine the initial second weights corresponding to each factor in the factor set based on feature information through an initial gating screening network; and convert the initial second weights corresponding to each factor into 0 / 1 through a preset activation function to obtain the second weights corresponding to each factor.

[0122] In some optional implementations of this embodiment, the sixth determining unit 704 is further configured to: multiply the first weight and the second weight corresponding to each factor in the factor set to obtain a weight product; and determine the target factor in the factor set that is suitable for the second user in the information flow recommendation scenario based on the weight product corresponding to each factor in the factor set.

[0123] In this embodiment, the target factor is determined based on the weight product of the first weight and the second weight corresponding to each factor, which further improves the convenience and accuracy of the target factor determination process during model training.

[0124] According to embodiments of this disclosure, this disclosure also provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to implement the information flow recommendation method and model training method described in any of the above embodiments.

[0125] According to embodiments of this disclosure, this disclosure also provides a readable storage medium storing computer instructions that enable a computer to implement the information flow recommendation method and model training method described in any of the above embodiments when executed.

[0126] This disclosure provides a computer program product that, when executed by a processor, can implement the information flow recommendation method and model training method described in any of the above embodiments.

[0127] Figure 8 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0128] like Figure 8 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0129] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0130] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as the information flow recommendation method. For example, in some embodiments, the information flow recommendation method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the information flow recommendation method described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to perform the information flow recommendation method by any other suitable means (e.g., by means of firmware).

[0131] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0132] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0133] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0134] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0135] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0136] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, also known as cloud computing servers or cloud hosts, which are hosting products within the cloud computing service system to address the management difficulties and weak business scalability inherent in traditional physical hosts and Virtual Private Servers (VPS) services; they can also be servers for distributed systems or servers integrated with blockchain technology.

[0137] According to the technical solution of the embodiments of this disclosure, an information flow recommendation method is provided. In the information flow recommendation scenario, a first weight of each factor corresponding to the user is determined by a multi-factor fusion parameter network, and a second weight of each factor is determined by a gating screening network. The target factors suitable for the user are accurately determined based on the first weight and the second weight, and information flow recommendation is performed, thereby improving the accuracy of the recommendation results.

[0138] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution provided in this disclosure can be achieved, and this is not limited herein.

[0139] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. An information flow recommendation method, comprising: Obtain the feature information of the first user in the information flow recommendation scenario; Through a multi-factor fusion parameter network, the first weight corresponding to each factor in the factor set is determined according to the feature information. The factors in the factor set represent the indicator information that needs to be considered in the information flow recommendation process. The factor set includes: reading time, number of user likes, and number of shares. The second weights corresponding to each factor in the factor set are determined by the gating network based on the feature information, wherein the second weights output by the gating network are 0 or 1. Multiply the first weight and the second weight corresponding to each factor in the factor set to obtain a weight product; determine the target factor in the factor set that is suitable for the first user in the information flow recommendation scenario based on the weight product corresponding to each factor in the factor set, wherein the factors with a weight product of 0 are removed, and the factors with a weight product of non-zero are retained as the target factors. Based on the target factor, determine and push the recommendation result corresponding to the first user in the information flow recommendation scenario to the first user; This also includes: Obtain feedback information from the first user regarding the recommendation results; Based on the feedback information and the preset reward function, the reward values ​​of the parameters of the multi-factor fusion parameter network and the parameters of the gated screening network are determined. Based on the principle of maximizing the reward value, the adjustment process of the parameters of the multi-factor fusion parameter network and the parameters of the gated screening network is guided. Based on the evolutionary strategy, the network is iterated according to a preset number of iterations to generate a new round of parameters of the multi-factor fusion parameter network and the parameters of the gated screening network that satisfy the Gaussian distribution of the mean and variance of each parameter. The adjusted multi-factor fusion parameter network and the gated screening network are then used to execute the recommendation task for subsequent users in the information flow recommendation scenario.

2. The method according to claim 1, wherein, The step of determining the second weight corresponding to each factor in the factor set based on the feature information through a gated screening network includes: The initial second weights corresponding to each factor in the factor set are determined using the gating network based on the feature information. The initial second weights of each factor are converted to 0 / 1 by a preset activation function, thus obtaining the second weights of each factor.

3. A model training method, comprising: Obtain the feature information of the second user in the information flow recommendation scenario; Through an initial multi-factor fusion parameter network, the first weight corresponding to each factor in the factor set is determined according to the feature information. The factors in the factor set represent the indicator information that needs to be considered in the information flow recommendation process. The factor set includes: reading time, number of user likes, and number of shares. The second weights corresponding to each factor in the factor set are determined by the initial gating network based on the feature information, wherein the second weights output by the initial gating network are 0 or 1. Multiply the first weight and the second weight corresponding to each factor in the factor set to obtain a weight product; determine the target factor in the factor set that is suitable for the second user in the information flow recommendation scenario based on the weight product corresponding to each factor in the factor set, wherein the factors with a weight product of 0 are removed, and the factors with a weight product of non-zero are retained as the target factors. Based on the target factor, determine the recommendation result corresponding to the second user in the information flow recommendation scenario; Based on the feedback information from the second user regarding the recommendation results and the preset reward function, the parameters of the multi-factor fusion parameter network and the parameters of the gated screening network are determined. Based on the principle of maximizing the reward value, the adjustment process of the parameters of the multi-factor fusion parameter network and the gated screening network is guided. Based on an evolutionary strategy, the network iterates according to a preset number of iterations to generate a new round of parameters for the multi-factor fusion parameter network and the gated screening network that satisfy a Gaussian distribution of the mean and variance of each parameter, thus obtaining the trained multi-factor fusion parameter network and the gated screening network.

4. The method according to claim 3, wherein, The step of determining the second weight corresponding to each factor in the factor set based on the feature information through an initial gating screening network includes: The initial second weights corresponding to each factor in the factor set are determined based on the feature information through the initial gating screening network. The initial second weights of each factor are converted to 0 / 1 by a preset activation function, thus obtaining the second weights of each factor.

5. An information flow recommendation device, comprising: The first acquisition unit is configured to acquire the feature information of the first user in the information flow recommendation scenario; The first determining unit is configured to determine the first weight corresponding to each factor in the factor set based on the feature information through a multi-factor fusion parameter network. The factors in the factor set represent the indicator information to be considered in the information flow recommendation process. The factor set includes: reading time, number of user likes, and number of shares. The second determining unit is configured to determine the second weight corresponding to each factor in the factor set based on the feature information through a gating screening network, wherein the second weight output by the gating screening network is 0 or 1; The third determining unit is configured to multiply the first weight and the second weight corresponding to each factor in the factor set to obtain a weight product; and to determine the target factor in the factor set that is applicable to the first user in the information flow recommendation scenario based on the weight product corresponding to each factor in the factor set, wherein the factors with a weight product of 0 are removed, and the factors with a weight product of non-zero are retained as the target factors. The recommendation unit is configured to determine and push the recommendation results corresponding to the first user in the information flow recommendation scenario to the first user based on the target factor; This also includes: The feedback unit is configured to obtain feedback information from the first user regarding the recommendation result; An evolutionary unit is configured to determine the return values ​​of the parameters of the multi-factor fusion parameter network and the parameters of the gated filtering network based on the feedback information and a preset return function; guide the adjustment process of the parameters of the multi-factor fusion parameter network and the gated filtering network based on the principle of maximizing the return value; iterate according to the evolutionary strategy for a preset number of iterations to generate a new round of parameters of the multi-factor fusion parameter network and the gated filtering network that satisfy the Gaussian distribution of the mean and variance of each parameter, so as to execute the recommendation task of subsequent users in the information flow recommendation scenario through the adjusted multi-factor fusion parameter network and the gated filtering network.

6. The apparatus according to claim 5, wherein, The second determining unit is further configured to: The gating network determines the initial second weights of each factor in the factor set based on the feature information; the initial second weights of each factor are converted to 0 / 1 using a preset activation function to obtain the second weights of each factor.

7. A model training device, comprising: The second acquisition unit is configured to acquire the feature information of the second user in the information flow recommendation scenario; The fourth determining unit is configured to determine the first weight corresponding to each factor in the factor set based on the feature information through an initial multi-factor fusion parameter network. The factors in the factor set represent the indicator information to be considered in the information flow recommendation process. The factor set includes: reading time, number of user likes, and number of shares. The fifth determining unit is configured to determine the second weight corresponding to each factor in the factor set based on the feature information through an initial gating screening network, wherein the second weight output by the initial gating screening network is 0 or 1; The sixth determining unit is configured to multiply the first weight and the second weight corresponding to each factor in the factor set to obtain a weight product; and to determine the target factor in the factor set that is applicable to the second user in the information flow recommendation scenario based on the weight product corresponding to each factor in the factor set, wherein the factors with a weight product of 0 are removed, and the factors with a weight product of non-zero are retained as the target factors. The seventh determining unit is configured to determine the recommendation result corresponding to the second user in the information flow recommendation scenario based on the target factor; The training unit is configured to determine the parameters of the multi-factor fusion parameter network and the parameters of the gated screening network based on the feedback information of the second user on the recommendation results and a preset reward function; guide the adjustment process of the parameters of the multi-factor fusion parameter network and the gated screening network based on the principle of maximizing the reward value; and iterate according to a preset number of iterations based on an evolutionary strategy to generate a new round of parameters of the multi-factor fusion parameter network and the gated screening network that satisfy the Gaussian distribution of the mean and variance of each parameter, so as to obtain the trained multi-factor fusion parameter network and the gated screening network.

8. The apparatus according to claim 7, wherein, The fifth determining unit is further configured to: The initial gating network is used to determine the initial second weights of each factor in the factor set based on the feature information; the initial second weights of each factor are converted to 0 / 1 by a preset activation function to obtain the second weights of each factor.

9. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-4.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-4.

11. A computer program product, comprising: A computer program that, when executed by a processor, implements the method according to any one of claims 1-4.

Images